Getting Started with Subtimizer

1. Overview

Subtimizer is a computational workflow for structure-guided design of potent and selective kinase peptide substrates. It integrates three state-of-the-art tools — AlphaFold-Multimer for structural modeling, ProteinMPNN for sequence design, and AlphaFold2-based interface evaluation — into an end-to-end pipeline for substrate engineering. The workflow automates SLURM job submission, sequence analysis, clustering, and validation, enabling high-throughput design and assessment of candidate substrates on HPC systems.

2. Prerequisites

Before installing Subtimizer, ensure that the following dependencies are available on your HPC system:

1. Anaconda, Miniconda, or Mamba — for conda environment management
2. ColabFold (with AlphaFold-Multimer support) — must be installed and added to $PATH
3. ProteinMPNN — install and set the MPNN_PATH environment variable to its directory
4. af2_initial_guess (from dl_binder_design) — install and set the DL_BINDER_DESIGN_PATH environment variable
5. SLURM-based HPC environment — for job scheduling and parallel execution

3. Installation

3.1. Create Conda Environments

Subtimizer requires separate conda environments to prevent dependency conflicts between AlphaFold and ProteinMPNN. First, create and activate the main subtimizer_env environment:

Step 1: Create and activate the main environment:

1
2

mamba create -n subtimizer_env python=3.9 -y
mamba activate subtimizer_env

Step 2: Create the worker environments from the YAML files included in the repository:

1
2

mamba env create -f af2_des_env.yaml
mamba env create -f mpnn_des_env.yaml

3.2. Install Subtimizer

With the subtimizer_env environment active, install Subtimizer either from PyPI (recommended) or directly from the source repository:

Option A: Install from PyPI (recommended)

1

pip install subtimizer

Option B: Install from source

1
2
3

git clone https://github.com/abeebyekeen/subtimizer.git
cd subtimizer
pip install .

Verify the installation:

1

subtimizer --help

4. Configuration

Subtimizer uses SLURM job templates to submit tasks to your HPC cluster. The default templates embedded in the package may use partition names, memory settings, or module load commands that differ from your cluster's configuration. To customize these, initialize local template files in your project directory:

1

subtimizer init-templates

This creates a subtimizer_templates/ directory with editable SLURM job scripts. Edit the partition names, memory allocations, and module requirements in these files to match your HPC environment. Subtimizer will use local templates from this directory in preference to the built-in defaults.

5. Usage

The Subtimizer workflow proceeds in twelve sequential steps, from initial structural modeling of the kinase–substrate complex through ProteinMPNN sequence design, clustering, and final validation. A list of complexes (one per line) is provided to each command via the --file flag. Most commands also accept --start and --end to process a subset of the list, and --max-jobs to control concurrency.

Step 1: Setup Project Structure

Initialize the project directory structure for your list of complexes:

1
2

cd examples
subtimizer setup --file example_list_of_complexes.dat --type initial

Step 2: Fold with AlphaFold-Multimer

Run AlphaFold-Multimer structural predictions for the kinase–substrate complexes. Two modes are available:

Batch mode (one job per complex, up to --max-jobs running simultaneously):

1

subtimizer fold --file example_list_of_complexes.dat --max-jobs 4

Parallel mode (all complexes in a single multi-GPU SLURM job):

1

subtimizer fold --file example_list_of_complexes.dat --mode parallel --max-jobs 4

Step 3: Design with ProteinMPNN

Set up ProteinMPNN input files, then run sequence design on the predicted structures:

1
2

subtimizer setup --file example_list_of_complexes.dat --type mpnn
subtimizer design --file example_list_of_complexes.dat --max-jobs 4

Step 4: Analyze Results

Parse and summarize the ProteinMPNN design outputs:

1

subtimizer analyze --file example_list_of_complexes.dat

Step 5: Cluster Sequences

Cluster the designed sequences to identify a diverse set of candidate substrates for downstream validation:

1

subtimizer cluster --file example_list_of_complexes.dat

Step 6: Prepare for Validation Folding

Prepare input files for folding the clustered (designed) sequences:

1

subtimizer prep-fold --file example_list_of_complexes.dat

Step 7: Fold Designed Sequences

Run AlphaFold-Multimer on the designed sequences for structural validation:

1

subtimizer fold --file example_list_of_complexes.dat --stage validation --max-jobs 4

Step 8: Fix PDBs

Renumber and clean up the output PDB files for downstream validation:

1

subtimizer fix-pdb --file example_list_of_complexes.dat

Step 9: Validate with AF2 Initial Guess

Validate the designed sequences using the AF2 initial guess methodology:

1

subtimizer validate --file example_list_of_complexes.dat --binder-path /path/to/predict.py

Replace /path/to/predict.py with the full path to the predict.py script from your dl_binder_design (af2_initial_guess) installation.

Step 10: Generate Reports

Generate summary reports and interface metrics for the designed sequences:

1

subtimizer report --file example_list_of_complexes.dat

Step 11: Parental Substrate Processing

Process and validate the original (parental) substrate for comparison with the designed sequences. This establishes a baseline for evaluating the improvement achieved by the designed candidates.

1
2
3
4
5
6

subtimizer setup --file example_list_of_complexes.dat --type original
cd original_subs
subtimizer fix-pdb --file ../example_list_of_complexes.dat
subtimizer validate --file ../example_list_of_complexes.dat --max-jobs 4
subtimizer report --file ../example_list_of_complexes.dat
cd ..

Then generate the final combined report comparing designed and parental substrates:

1

subtimizer report --file example_list_of_complexes.dat

Step 12: ipSAE Evaluation (Optional)

ipSAE provides an additional interface quality metric based on predicted aligned error (PAE). This step is optional but recommended for a comprehensive evaluation of the designed substrates.

First, download ipSAE and add it to your $PATH:

1

export PATH=$PATH:/path/to/ipSAE_directory

Then run the ipSAE evaluation:

1

subtimizer ipsae --file example_list_of_complexes.dat --max-jobs 8

6. Citation

If you use Subtimizer in your research, please cite:

Yekeen A.A., Meyer C.J., McCoy M., Posner B., Westover K.D. Subtimizer: Computational Workflow for Structure-Guided Design of Potent and Selective Kinase Peptide Substrates. Journal of Chemical Information and Modeling (2026). GitHub .

I try my best to make the information on this website as accurate as possible, but if you find any errors in the contents of this page or any other page on this website, kindly get in touch with me at contact[at]abeebyekeen[dot]com. Also, you are welcome to reach out for assistance and collaboration.